MyRoboPath
computer vision17 min readUpdated 2026-03-06Advanced

Visual Odometry (VO) with ORB Features & Epipolar Geometry

Estimate camera 6-DOF trajectory from monocular and stereo video frames using Oriented FAST and Rotated BRIEF (ORB) feature matching and Essential Matrix decomposition.

Dr. Elena Rostova
Dr. Elena Rostova
Principal Computer Vision Scientist

Key Engineering Takeaways

  • Visual Odometry estimates frame-to-frame incremental camera poses without building a global persistent map or running loop closure.
  • The Essential Matrix E = [t]ₓ R connects corresponding normalized image coordinates: x₂ᵀ E x₁ = 0.
  • Decomposing E via Singular Value Decomposition (SVD) produces 4 possible [R|t] solutions; cheirality check selects the one where 3D points lie in front of both cameras.
Prerequisites
  • Linear algebra
  • OpenCV feature detectors

Computing Essential Matrix E & Recovering [R|t]

Epipolar geometry governs the relationship between two views. Using the 5-point or 8-point algorithm inside RANSAC, the Essential Matrix $\mathbf{E}$ is computed and decomposed into rotation $\mathbf{R} \in SO(3)$ and unit translation $\mathbf{t}$.
Tags:#Visual Odometry#ORB#SLAM#Epipolar Geometry#Essential Matrix#Feature Matching